3-D video depth map quantization based on Lloyd's algorithm

Abstract

Depth map based 3-D view synthesis technique allows to synthesize different virtual perspectives of the same 3-D content, and the lifelike 3-D view perception of these virtual perspectives is highly depending on the accuracy of depth information. So, although depth map is not viewed by the users, however, its quality directly affects that of the synthesized virtual views by the 3-D warping process. In view of current progress in depth map acquisition techniques, and in order to maintain compatibility with different encoding standards depth values need to undergo quantization, so as to be represented at practical and convenient values. However, the quantization process causes depth map distortion. In order to reduce this negative effect, this paper propose a new quantization method which jointly considers the distribution of depth values and the 3-D rendering process, so as to minimize the depth map induced rendering distortion. The experimental results with the proposed quantization method indicate that the SNR gain has increased from 1dB to 14dB for different sequences when compared with the traditional non-uniform quantization method.